[RVV] Moved weight repacking for FC to the gemm

openvinotoolkit · Jul 11, 2024 · 3a7d230 · 3a7d230
1 parent 9fb8aac
commit 3a7d230
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 10 deletions.
diff --git a/source/thead_rvv/fp32/fullyconnected.c b/source/thead_rvv/fp32/fullyconnected.c
@@ -22,16 +22,7 @@ int shl_rvv_fullyconnected_init_fp32(struct csinn_tensor *input, struct csinn_te
                                      struct csinn_tensor *weights, struct csinn_tensor *bias,
                                      struct csinn_fc_params *params)
 {
-    const int weights_dims_count = weights->dim_count;
-    const int out_nodes = weights->dim[weights_dims_count - 2];
-    const int in_nodes = weights->dim[weights_dims_count - 1];
     struct csinn_callback *cb = params->base.cb;
-
-    struct csinn_session *sess = params->base.sess;
-    bool binary_model_op_init = shl_rvv_get_binary_model_op_init(sess);
-    if (!binary_model_op_init) {
-        shl_rvv_fc_gemm_reorder_weight_fp32(weights);
-    }
     cb->exec = shl_rvv_fullyconnected_gemm_fp32;
 
     return CSINN_TRUE;

diff --git a/source/thead_rvv/fp32/fullyconnected_fp32.c b/source/thead_rvv/fp32/fullyconnected_fp32.c
@@ -93,9 +93,12 @@ int shl_rvv_fullyconnected_gemm_fp32(struct csinn_tensor *input, struct csinn_te
     int k = accum_depth;
 
     float *input_reorder = (float *)shl_mem_alloc(m * k * sizeof(float));
+    float *weights_reorder = (float *)shl_mem_alloc(n * k * sizeof(float));
+    reorder_weight_npack2n_fp32(weights_data, weights_reorder, n, k);
     shl_rvv_reorder_a_block_12xk_fp32(input_data, input_reorder, m, k, m, k);
-    shl_rvv_gemm_a0b1_12xpack2n_fp32(output_data, input_reorder, weights_data, bias_data, m, k, n);
+    shl_rvv_gemm_a0b1_12xpack2n_fp32(output_data, input_reorder, weights_reorder, bias_data, m, k, n);
 
     shl_mem_free(input_reorder);
+    shl_mem_free(weights_reorder);
     return CSINN_TRUE;
 }